Publication number | US20040205038 A1 |

Publication type | Application |

Application number | US 10/411,437 |

Publication date | Oct 14, 2004 |

Filing date | Apr 9, 2003 |

Priority date | Apr 9, 2003 |

Also published as | US7263509 |

Publication number | 10411437, 411437, US 2004/0205038 A1, US 2004/205038 A1, US 20040205038 A1, US 20040205038A1, US 2004205038 A1, US 2004205038A1, US-A1-20040205038, US-A1-2004205038, US2004/0205038A1, US2004/205038A1, US20040205038 A1, US20040205038A1, US2004205038 A1, US2004205038A1 |

Inventors | Shih-Jong Lee, Seho Oh |

Original Assignee | Lee Shih-Jong J., Seho Oh |

Export Citation | BiBTeX, EndNote, RefMan |

Patent Citations (5), Referenced by (7), Classifications (15), Legal Events (5) | |

External Links: USPTO, USPTO Assignment, Espacenet | |

US 20040205038 A1

Abstract

An intelligent spatial reasoning method receives a plurality of object sets. A spatial mapping feature learning method uses the plurality of object sets to create at least one salient spatial mapping feature output. It performs spatial reasoning rule learning using the at least one spatial mapping feature to create at least one spatial reasoning rule output. The spatial mapping feature learning method performs a spatial mapping feature set generation step followed by a feature learning step. The spatial mapping feature set is generated by repeated application of spatial correlation between two object sets. The feature learning method consists of a feature selection step and a feature transformation step and the spatial reasoning rule learning method uses the supervised learning method.

The spatial reasoning approach of this invention automatically characterizes spatial relations of multiple sets of objects by comprehensive collections of spatial mapping features. Some of the features have clearly understandable physical, structural, or geometrical meanings. Others are statistical characterizations, which may not have clear physical, structural or geometrical meanings when considered individually. A combination of these features, however, could characterize subtle physical, structural or geometrical conditions under practical situations. One key advantage of this invention is the ability to characterize subtle differences numerically using a comprehensive feature set.

Claims(35)

a) Input a plurality of object sets;

b) Perform spatial mapping feature learning using the plurality of object sets to create at least one salient spatial mapping feature output;

c) Perform spatial reasoning rule learning using the at least one spatial mapping feature to create at least one spatial reasoning rule output.

a) Perform spatial mapping feature set generation using the plurality of object sets to create a spatial mapping feature set output;

b) Perform feature learning using the spatial mapping feature set to create at least one salient spatial mapping feature output.

a) Input a target object set;

b) Input a condition object set;

c) Perform label propagation operation on the target object set and the condition object set to create target object set transformed data and condition object set transformed data output;

d) Perform a spatial mapping features calculation using the target object set transformed data and the condition object set transformed data to create a spatial mapping features output.

a) No operation,

b) Inner distance transformation,

c) Outer distance transformation,

d) Connected component labeling, and

e) Zone of influence labeling.

a) Condition ratio, and

b) Intersection ratio.

a) Inner distance mean,

b) Inner distance standard deviation,

c) Normalized inner distance mean,

d) Normalized inner distance standard deviation,

e) Inner distance skewness, and

f) Inner distance kurtosis.

a) Outer distance mean,

b) Outer distance standard deviation,

c) Normalized outer distance mean;

d) Normalized outer distance standard deviation,

e) Outer distance skewness, and

f) Outer distance kurtosis.

a) Intersection-component-count,

b) Intersection-component-count-ratio,

c) Intersection-component-average-area,

d) Intersection-component-sd-area,

e) Intersection-component-average-area-ratio,

f) Intersection-component-sd-area-ratio, and

g) Intersection-component-area-entropy.

a) ZOI_intersection-component-count,

b) ZOI_intersection-component-count-ratio,

c) ZOI_intersection-component-average-area,

d) ZOI_intersection-component-sd-area,

e) ZOI_intersection-component-average-area-ratio,

f) ZOI_intersection-component-sd-area-ratio,

g) ZOI_intersection-component-area-entropy, and

h) Intersection-component-count_proportion.

a) Target-inner-distance-mean,

b) Target-inner-distance-standard-deviation,

c) Normalized-target-inner-distance-mean,

d) Normalized-target-inner-distance-standard-deviation,

e) Target-inner-distance-skewness, and

f) Target-inner-distance-kurtosis.

a) P_{I} _{ — } _{ID}—Energy,

b) P_{I} _{ — } _{ID}—Entropy,

c) P_{I} _{ — } _{ID}—Correlation,

d) P_{I} _{ — } _{ID}—Inertia,

e) P_{I} _{ — } _{ID}—Homogeneity,

f) P_{I} _{ — } _{ID}—Max_probability,

g) P_{I} _{ — } _{ID}—Cluster_tendency, and

h) P_{I} _{ — } _{ID}—Deep_overlap_tendency.

a) P_{I} _{ — } _{OD}—Energy,

b) P_{I} _{ — } _{OD}—Entropy,

c) P_{I} _{ — } _{OD}—Correlation,

d) P_{I} _{ — } _{OD}—Inertia,

e) P_{I} _{ — } _{OD}—Homogeneity,

f) P_{I} _{ — } _{OD}—Max_probability,

g) P_{I} _{ — } _{OD}—Cluster_tendency, and

h) P_{I} _{ — } _{OD}—Deep_overlap_tendency.

a) Target-inner-distance-mean-standard-deviation,

b) Target-inner-distance-standard-deviation_mean,

c) Target-inner-distance-standard-deviation_sd,

d) Normalized-target-inner-distance-mean-standard-deviation,

e) Normalized-target-inner-distance-standard-deviation_mean, and

f) Normalized-target-inner-distance-standard-deviation_sd.

a) ZOI_target-inner-distance-mean-standard-deviation,

b) ZOI_target-inner-distance-standard-deviation_mean,

c) ZOI_target-inner-distance-standard-deviation_sd,

d) ZOI_normalized-target-inner-distance-mean-standard-deviation,

e) ZOI_normalized-target-inner-distance-standard-deviation_mean, and

f) ZOI_normalized-target-inner-distance-standard-deviation_sd.

a) Target-outer-distance-mean,

b) Target-outer-distance-standard-deviation,

c) Normalized-target-outer-distance-mean,

d) Normalized-target-outer-distance-standard-deviation,

e) Target-outer-distance-skewness, and

f) Target-outer-distance-kurtosis.

a) P_{O} _{ — } _{ID}—Energy,

b) P_{O} _{ — } _{ID}—Entropy,

c) P_{O} _{ — } _{ID}—Correlation,

d) P_{O} _{ — } _{ID}—Inertia,

e) P_{O} _{ — } _{ID}—Homogeneity,

f) P_{O} _{ — } _{ID}—Max_probability,

g) P_{O} _{ — } _{ID}—Cluster_tendency, and

h) P_{O} _{ — } _{ID}—Deep_overlap_tendency.

a) P_{O} _{ — } _{OD}—Energy,

b) P_{O} _{ — } _{OD}—Entropy,

c) P_{O} _{ — } _{OD}—Correlation,

d) P_{O} _{ — } _{OD}—Inertia,

e) P_{O} _{ — } _{OD}—Homogeneity,

f) P_{O} _{ — } _{OD}—Max_probability,

g) P_{O} _{ — } _{OD}—Cluster_tendency, and

h) P_{O} _{ — } _{OD}—Inside_overlap_tendency.

a) Target-outer-distance-mean-standard-deviation,

b) Target-outer-distance-standard-deviation_mean,

c) Target-outer-distance-standard-deviation_sd,

d) Normalized-target-outer-distance-mean-standard-deviation,

e) Normalized-target-outer-distance-standard-deviation_mean, and

f) Normalized-target-outer-distance-standard-deviation_sd.

a) ZOI_target-outer-distance-mean-standard-deviation,

b) ZOI_target-outer-distance-standard-deviation_mean,

c) ZOI_target-outer-distance-standard-deviation_sd,

d) ZOI_normalized-target-outer-distance-mean-standard-deviation,

e) ZOI_normalized-target-outer-distance-standard-deviation_mean, and

f) ZOI_normalized-target-outer-distance-standard-deviation_sd.

a) Intersection-target-component-count,

b) Intersection-target-component-count-ratio,

c) Intersection-target-component-average-area,

d) Intersection-target-component-sd-area,

e) Intersection-target-component-average-area-ratio,

f) Intersection-target-component-sd-area-ratio, and

g) Intersection-target-component-area-entropy.

a) Condition-inner-distance-mean-standard-deviation,

b) Condition-inner-distance-standard-deviation_mean,

c) Condition-inner-distance-standard-deviation_sd,

d) Normalized-condition-inner-distance-mean-standard-deviation,

e) Normalized-condition-inner-distance-standard-deviation_mean, and

f) Normalized-condition-inner-distance-standard-deviation_sd.

a) Condition-outer-distance-mean-standard-deviation,

b) Condition-outer-distance-standard-deviation_mean,

c) Condition-outer-distance-standard-deviation_sd,

d) Normalized-condition-outer-distance-mean-standard-deviation,

e) Normalized-condition-outer-distance-standard-deviation_mean, and

f) Normalized-condition-outer-distance-standard-deviation_sd.

a) Target_intersection-component-count_mean,

b) Target_intersection-component-count_sd,

c) Target_intersection-component-count-ratio_mean,

d) Target_intersection-component-count-ratio_sd,

e) Target_intersection-component-average-area_mean,

f) Target_intersection-component-average-area_sd,

g) Target_intersection-component-average-area-ratio_mean,

h) Target_intersection-component-average-area-ratio_sd,

i) Target_intersection-component-area-entropy_mean, and

j) Target_intersection-component-area-entropy_sd.

a) Target_intersection-ZOI-component-count_mean,

b) Target_intersection-ZOI-component-count_sd,

c) Target_intersection-ZOI-component-count-ratio_mean,

d) Target_intersection-ZOI-component-count-ratio_sd,

e) Target_intersection-ZOI-component-area_mean,

f) Target_intersection-ZOI-component-average-area_sd,

g) Target_intersection-ZOI-component-average-area-ratio_mean,

h) Target_intersection-ZOI-component-average-area-ratio_sd,

i) Target_intersection-ZOI-component-area-entropy_mean, and

j) Target_intersection-ZOI-component-area-entropy_sd.

a) Intersection-target-ZOI-component-count,

b) Intersection-target-ZOI-component-count-ratio,

c) Intersection-target-ZOI-component-average-area,

d) Intersection-target-ZOI-component-sd-area,

e) Intersection-target-ZOI-component-average-area-ratio,

f) Intersection-target-ZOI-component-sd-area-ratio, and

g) Intersection-target-ZOI-component-area-entropy.

a) Condition-inner-distance-mean-ZOI-standard-deviation,

b) Condition-inner-distance-standard-deviation-ZOI-mean,

c) Condition-inner-distance-standard-deviation-ZOI_sd,

d) Normalized-condition-inner-distance-mean-ZOI-standard-deviation,

e) Normalized-condition-inner-distance-standard-deviation-ZOI_mean, and

f) Normalized-condition-inner-distance-standard-deviation-ZOI_sd.

a) Condition-outer-distance-mean-ZOI-standard-deviation,

b) Condition-outer-distance-standard-deviation-ZOI-mean,

c) Condition-outer-distance-standard-deviation-ZOI_sd,

d) Normalized-condition-outer-distance-mean-ZOI-standard-deviation,

e) Normalized-condition-outer-distance-standard-deviation-ZOI_mean, and

f) Normalized-condition-outer-distance-standard-deviation-ZOI_sd.

a) Target-ZOI_intersection-component-count_mean,

b) Target-ZOI_intersection-component-count_sd,

c) Target-ZOI_intersection-component-count-ratio_mean,

d) Target-ZOI_intersection-component-count-ratio_sd,

e) Target-ZOI_intersection-component-area_mean,

f) Target-ZOI_intersection-component-average-area_sd,

g) Target-ZOI_intersection-component-average-area-ratio_mean,

h) Target-ZOI_intersection-component-average-area-ratio_sd,

i) Target-ZOI_intersection-component-area-entropy_mean, and

j) Target-ZOI_intersection-component-area-entropy_sd.

a) Target-ZOI_intersection-ZOI-component-count_mean,

b) Target-ZOI_intersection-ZOI-component-count_sd,

c) Target-ZOI_intersection-ZOI-component-count-ratio_mean,

d) Target-ZOI_intersection-ZOI-component-count-ratio_sd,

e) Target-ZOI_intersection-ZOI-component-area_mean,

f) Target-ZOI_intersection-ZOI-component-average-area_sd,

g) Target-ZOI_intersection-ZOI-component-average-area-ratio_mean,

h) Target-ZOI_intersection-ZOI-component-average-area-ratio_sd,

i) Target-ZOI_intersection-ZOI-component-area-entropy_mean, and

j) Target-ZOI_intersection-ZOI-component-area-entropy_sd.

a) Input a target object set;

b) Input a plurality of condition object sets;

c) Perform spatial mapping features extraction for the target object set and each of the plurality of condition object sets to create spatial mapping features outputs for each dual of the object set pairs;

d) Perform contrast feature extraction using spatial mapping features for each dual of the object set pairs to create a contrast features output.

a) Feature difference,

b) Absolute feature difference,

c) First feature proportion,

d) Second feature proportion, and

e) Absolute difference proportion.

Description

- [0001]This invention relates to a method for intelligent spatial reasoning of multiple sets of objects in multidimensional space.
- [0002]Spatial reasoning is commonly known as reasoning about shape, measurement, depiction and navigation. Shape relates to the structure of space. Measurement relates to the dimension of space. Depiction relates to the representation of space entities. Navigation relates to large-scale space and concepts such as position, direction, distance, and routes. Spatial reasoning creates representation and applies rules that could describe the relations or changes in the relations of shape, measurement, depiction and navigation.
- [0003]Humans develop spatial reasoning capability by first developing spatial representation then integrating multiple representations by their associated relations and rules. Similarly, spatial reasoning by computer involves the development of models for representation of spatial entities and the inference of information about spatial relations within the model framework. Spatial knowledge could be stored in a human cognitive map like representation such as the Spatial Semantic Hierarchy (B. Kuipers. 2000
*. The Spatial Semantic Hierarchy. Artificial Intelligence*119: 191-233.) The prevailing goal of a computerized spatial reasoning method is to discover meaningful cues and premises from the spatial relations of multiple sets of objects residing in multidimensional space. - [0004]There are many practical applications for spatial reasoning. For example, spatial reasoning is critical to the Geographical Information Systems (GIS). GIS is a computer system capable of assembling, storing, manipulating, and displaying geographically referenced information, i.e. data identified according to its location. It can be used for scientific investigations, resource management, and development planning. (Geographical Information Systems http://www.usgs.gov/research/gis/title.html)
- [0005]In a cell image informatics application, spatial reasoning relates the target segmentation to specific structures and objects identified by morphology processing. For example in a Cajal Bodies (CB) dynamics study (Platani M., Goldberg I., Lamond A., Swedlow S. “Cajal body dynamics and association with chromatin are ATP-dependent”.
*Nature Cell Biology*. 2002 July; 4: 502-508), CBs identified in the target processing can be spatially related to the chromatin structures identified in the morphology processing. In semiconductor or electronic automatic defect classification or military automatic target classification applications, spatial reasoning relates the spatial relations among the components of defect arrangements or target structures to arrive at a classification decision. - [0006]In practical applications involving spatial decisions, it is highly desirable to have an automatic method to extract spatial reasoning features and generate spatial reasoning rules that could distinguish subtle differences and can compensate for measurement imperfection, noise, and uncertainty.
- [0007]Prior art approaches use symbolic representations such as wire frame, surface-edge-vertex, generalized cylinder, superquadric, octree, etc. for spatial entity representations (Haralick, R M, Shapiro, R G, “Computer and Robot Vision”, Vol. II, PP440-453, Addison Wesley, 1993). Relational distance based matching is often used for spatial reasoning based on symbolic representations. This class of methods suffers from the non-numerical nature of the representation and data processing where it is difficult to handle uncertainty and noise from data and measurements. Furthermore, it is difficult to automatically generate spatial mapping features and automatically generate or discover spatial mapping rules corresponding to events or targets of interest. Therefore, spatial reasoning rules are often defined by human heuristics. Human defined rules are primitive, i.e. non-robust. They are incapable of distinguishing subtle differences; neither can they compensate for imperfect measurements.
- [0008]The primary object of intelligent spatial reasoning is to automatically discover meaningful cues and premises from the spatial relations of multiple sets of objects residing in multidimensional space. A further object of this invention is to create spatial reasoning rules that are robust to practical situations and real life variation. A still further object of the invention is to automatically generate salient spatial mapping features to characterize the spatial situations of interest. A further object of the invention is to automatically generate spatial reasoning rules to identify and represent spatial situations of interest.
- [0009]The spatial reasoning approach of the invention automatically characterizes spatial relations of multiple sets of objects by comprehensive collections of spatial mapping features. Some of the features have clearly understandable physical, structural, or geometrical meanings. Others are statistical characterizations, which may not have clear physical, structural or geometrical meanings when considered individually. A combination of these features, however, could characterize subtle physical, structural or geometrical conditions under practical situations. One key advantage of the invention is the ability to characterize subtle differences numerically using a comprehensive feature set.
- [0010]Comprehensive numerical characterization facilitates the effectiveness of the follow on feature learning stage to automatically select salient features among the comprehensive collection of features. The feature learning process optimizes the discrimination power of features by feature selection, feature combination or feature transformation. Since the features are derived from measured data that include all imperfection, noise, variations, and measurement errors, the feature learning process embraces all non-ideal yet practical situations and the learned features are robust to real life variations.
- [0011]After the feature learning method of this invention automatically selects salient features, spatial reasoning rules are automatically generated by decision learning methods. The spatial reasoning rule learning stage generates rules that include a combination of the salient features to arrive at physical, structural, or geometrical conditions that are strongly related to the events/outcomes/classification of interest. The spatial reasoning rules can be presented to a human for confirmation and adjustment if needed. The resulting spatial reasoning rules can be applied to new data for automatic decision making from spatial relations. That is, the resulting spatial reasoning rules can be used to perform automatic spatial reasoning on volume data or new experiments.
- [0012]An intelligent spatial reasoning method receives a plurality of object sets. A spatial mapping feature learning method uses the plurality of object sets to create at least one salient spatial mapping feature output. It performs spatial reasoning rule learning using the at least one spatial mapping feature to create at least one spatial reasoning rule output. The spatial mapping feature learning method performs a spatial mapping feature set generation step followed by a feature learning step. The spatial mapping feature set is generated by repeated application of spatial correlation between two object sets. The feature learning method consists of a feature selection step and a feature transformation step and the spatial reasoning rule learning method uses the supervised learning method.
- [0013]The preferred embodiment and other aspects of the invention will become apparent from the following detailed description of the invention when read in conjunction with the accompanying drawings which are provided for the purpose of describing embodiments of the invention and not for limiting same, in which:
- [0014][0014]FIG. 1 shows the overall processing flow for the spatial reasoning method;
- [0015][0015]FIG. 2 shows the processing flow for the spatial mapping feature learning method;
- [0016][0016]FIG. 3 shows the processing flow for the spatial mapping features generation method for a target object set and a condition object set;
- [0017][0017]FIG. 4A shows an example image of an object set with 4 squares;
- [0018][0018]FIG. 4B shows the 8-connected inner distance transformed image for the image of FIG. 4A;
- [0019][0019]FIG. 4C shows the 4-connected outer distance transformed image for the image of FIG. 4A;
- [0020][0020]FIG. 4D shows the connected component labeled image for the image of FIG. 4A;
- [0021][0021]FIG. 4E shows the zone of influence labeled image for the image of FIG. 4A;
- [0022][0022]FIG. 5A shows an example image of an object set with one circle region;
- [0023][0023]FIG. 5B shows the 8-connected inner distance transformed image for the image of FIG. 5A;
- [0024][0024]FIG. 5C shows the 4-connected outer distance transformed image for the image of FIG. 5A;
- [0025][0025]FIG. 5D shows the connected component labeled image for the image of FIG. 5A;
- [0026][0026]FIG. 5E shows the zone of influence labeled image for the image of FIG. 5A;
- [0027][0027]FIG. 6A shows an example image of an object set with one non-convex region formed by two touching circles;
- [0028][0028]FIG. 6B shows inner distance transformed image for the image of FIG. 6A;
- [0029][0029]FIG. 6C shows local centers for the image of FIG. 6A;
- [0030][0030]FIG. 6D shows local center distance transformed image for the image of FIG. 6A;
- [0031][0031]FIG. 7 lists the feature collections and the primary information they extract;
- [0032][0032]FIG. 8 shows the processing flow for spatial mapping feature generalization for multiple object sets;
- [0033][0033]FIG. 9A shows the boundary of the object set of FIG. 5A (as no operation target object set) overlaid on the image of FIG. 4A (as no operation condition object set). This illustrates the underlying spatial relation between the no operation target object set and the no operation condition object set;
- [0034][0034]FIG. 9B shows the boundary of the object set of FIG. 5A (as no operation target object set) overlaid on the image of FIG. 4B (as inner distance transformed condition object set). This illustrates the underlying spatial relation between the no operation target object set and the inner distance transformed condition object set;
- [0035][0035]FIG. 9C shows the boundary of the object set of FIG. 5A (as no operation target object set) overlaid on the image of FIG. 4C (as outer distance transformed condition object set). This illustrates the underlying spatial relation between the no operation target object set and the outer distance transformed condition object set;
- [0036][0036]FIG. 9D shows the boundary of the object set of FIG. 5A (as no operation target object set) overlaid on the image of FIG. 4D (as connected component labeled condition object set). This illustrates the underlying spatial relation between the no operation target object set and the connected component labeled condition object set;
- [0037][0037]FIG. 9E shows the boundary of the object set of FIG. 5A (as no operation target object set) overlaid on the image of FIG. 4E (as zone of influence labeled condition object set). This illustrates the underlying spatial relation between the no operation target object set and the zone of influence labeled condition object set.
- [0038]The overall processing flow of the spatial reasoning process is shown in FIG. 1. A plurality of object sets
**104**,**106**, . . . ,**108**is processed by a spatial mapping feature learning step**100**. This step generates a set of salient spatial mapping features**110**that are processed by the follow on spatial reasoning rule learning step**102**. The spatial reasoning rule learning step automatically creates spatial reasoning rules**112**that characterize the spatial situations of interest among the plurality of object sets**104**,**106**, . . .**108**. - [0039]The spatial mapping feature learning step generates a comprehensive set of spatial mapping features. The features are subject to a feature learning process to transform and select a set of salient spatial mapping features for follow-on processing. This is shown in FIG. 2. A spatial mapping feature set generation method
**200**uses the plurality of object sets**104**,**106**, . . . ,**108**to create a spatial mapping feature set**204**that includes a plurality of features. The spatial mapping feature set**204**is processed by a feature learning step**202**to create at least one salient spatial mapping feature**110**. - [0040]II.1 Spatial Mapping Feature Set Generation
- [0041]The spatial mapping feature set is generated by repeated application of spatial correlation between two object sets. We first describe the procedure of spatial mapping feature generation between two object sets: object set
**1**and object set**2**. - [0042]The rationale behind the spatial mapping feature set generation is to create a comprehensive characterization of all possible spatial relations between the object sets. The comprehensive feature set facilitates the effectiveness of the follow-on feature learning stage to select salient features from the set. In addition, a spatial reasoning rule learning stage generates rules that use the combination of salient features and decision criteria to arrive at physical, structural, geometrical, or other conditions that may be related to the events/conditions/classifications of interest underlying the object sets.
- [0043]II.1.1 Spatial Mapping Features Between Two Object Sets
- [0044]In one embodiment of the invention, binary masks in a multidimensional image space represent each object set. The pixels belonging to the object set are labeled “1”, otherwise the pixels are labeled “0”. To generate the spatial mapping features by object set correlation, one of the object sets is designated as the target object set and the other object set is designated as the condition object set.
- [0045]Each object set is subject to transformation by several label propagation operations. Spatial mapping features for the target object set are calculated from the transformed images of the target object set and the condition object set. The target and condition object set designation can be reversed to derive another set of spatial mapping features. This would enrich the spatial characterization power of the spatial mapping features. The processing flow for the spatial mapping features generation method for a target object set and a condition object set is shown in FIG. 3. The spatial mapping feature generation method inputs a target object set
**304**and a condition object set**306**. It performs label propagation operation**300**on the target object set**304**and the condition object set**306**to create target object set transformed data**308**and condition object set transformed data**310**outputs. It then performs spatial mapping features calculation**302**using the target object set transformed data**308**and the condition object set transformed data**310**. This results in a spatial mapping features output**312**. - [0046]II.1.1.1 Label Propagation Operations
- [0047]In one embodiment of the invention, 5 operations are applied: no operation, inner distance transformation, outer distance transformation, connected component labeling, zone of influence labeling.
- [0048]Without loss of generality, we explain the operations for 2 object sets in two-dimensional images. The method can be directly generalized to multidimensional images. Other dimensions could include color, time, momentum, intensity, etc.
- [0049]In this case, the input binary image is used directly without any additional operation. The input binary image has a single label for each pixel belonging to the object set. FIG. 4A shows an example image whose object set includes 4 squares. The pixels belonging to the object set are labeled “255” (brightest regions).
- [0050]Inner distance transformation creates a distance image that assigns the shortest distance from each of the object pixels to a background pixel. In a two dimensional image, the distance transformation can be performed efficiently by a two-pass algorithm. The first pass scans the image line by line from upper left corner toward bottom right corner and the second pass reverses the scanning from bottom right corner toward upper left corner. It uses a dynamic programming like optimization algorithm. (Rosenfeld A., Kak, A C., “Digital Picture Processing,” Academic Press, 1976, pp.356-357) The algorithm can be directly generalized to multidimensional images. FIG. 4B shows the inner distance transformed image of FIG. 4A where brighter pixels have higher distance values.
- [0051]Outer distance transformation creates a distance image that assigns the shortest distance from each of the background pixels to an object pixel. This can be achieved by applying the distance transformation on the background rather than the object regions. FIG. 4C shows the outer distance transformed image of FIG. 4A.
- [0052]A connected component labeling operation groups object pixels into connected components. Each object pixel is given a label corresponding to its connected component number. The object set shown in FIG. 4A has four connected components, one per square region. FIG. 4D shows a connected component labeled image of FIG. 4A where different labels are illustrated by different brightness. In one embodiment of the invention, the connected component labeling is performed by a two-pass algorithm (Haralick, R M, Shapiro, R G, “Computer and Robot Vision”, Vol. I, PP28-48, Addison Wesley, 1992) that can be directly generalized to multidimensional images.
- [0053]A zone of influence operation assigns each pixel of the image a unique label corresponding to the connected component to which it is closest (influenced). It can be determined by first finding the ridge lines (medial axis) from the outer distance transformed image (Serra, J, “Image analysis and mathematical morphology,” London: Academic Press, 1982, pp. 381-387) and performing connected component labeling of the non-ridge lines. FIG. 4E shows the zone of influence labeled image of FIG. 4A. In this image the image border is also considered a connected component.
- [0054][0054]FIG. 5 shows another example image whose object set includes a single circle region. The pixels belonging to the object set are labeled “255” (brightest regions) as shown in FIG. 5A. The inner distance transformed image (using 8-connected distance) is shown in FIG. 5B. The outer distance transformed image (using 4-connected distance) is shown in FIG. 5C. The connected component labeled image is shown in FIG. 5D and the zone of influence labeled image is shown in FIG. 5E.
- [0055]The label propagation operations are applied to both the target and the condition object set. This creates target object set transformed data and condition object set transformed data. The data are used to perform a spatial mapping features calculation.
- [0056]Those to create ordinary skill in the art should recognize that other label propagation operations or spatial transformations could be used and they are within the scope of this invention. An alternative transformation example is shown in FIG. 6D. FIG. 6A shows an example image of an object set with one non-convex region formed by two touching circles. FIG. 6B shows the inner distance transformed image for the image of FIG. 6A. FIG. 6C shows local centers for the image of FIG. 6A. Where local centers are the centers of local convex components of the object. They can be detected as the peaks in the inner distance transformed image. FIG. 6D shows the local center distance transformed image for the image of FIG. 6A.
- [0057]II.1.1.2 Spatial Mapping Features Calculation
- [0058]In one embodiment of the invention, there are up to 25 collections of features in the comprehensive spatial mapping feature set. The feature collections and the primary information extracted are listed in FIG. 7.
- [0059]Information from multiple aspects is extracted by the 25 collections of spatial mapping features. The definitions of the spatial mapping features are listed in section IV. The previous section describes the spatial mapping feature set between two object sets when one of them is designated as the target object set. The designation of the target object set and the condition object set can be reversed and the spatial mapping features recalculated even though some of the features would have already been extracted regardless of the target and condition object set designation. Some additional features can be derived by designation reversal for feature learning.
- [0060]Those having ordinary skill in the art should recognize that other forms of spatial mapping feature calculations could be used and they are within the scope of the invention.
- [0061]II.1.2 Spatial Mapping Features Between Multiple Object Sets
- [0062]The spatial mapping feature calculations for two object sets could be generalized to include multiple object sets. In one embodiment of the invention, the generalization method is shown in FIG. 8. The target object set
**806**is paired with each of the other object sets**808**,**810**one by one and the spatial mapping features**812**of each object set pair are extracted using the spatial mapping feature extraction method**800**,**802**as described in the previous section. Multiple feature sets created from two object set pairs could be further processed to create higher order contrast features**814**. The contrast features**814**can be extracted by the contrast feature extraction method**804**from each dual of the object set pairs. - [0063]Contrast features can be derived for each of the spatial mapping features. Assuming spatial mapping feature i is derived from the target object set and the object set
**1**(Feature_{i}^{T}^{ — }^{O1}) and is also derived from the target object set and the object set**2**(Feature_{i}^{T}^{ — }^{O2}) The contrast feature i for the (target, object set**1**) and (target, object set**2**) dual can be derived as: - [0064]Feature difference: F1
_{i}^{T}^{ — }^{O1-T}^{ — }^{02}=Feature_{i}^{T}^{ — }^{O1}−Feature_{i}^{T}^{ — }^{O2}, - [0065]Absolute feature difference: F2
_{i}^{T}^{ — }^{O1-T}^{ — }^{02}=|Feature_{i}^{T}^{ — }^{O1}−Feature_{i}^{T}^{ — }^{O2}|, - [0066]First feature proportion: F3
_{i}^{T}^{ — }^{O1-T}^{ — }^{02}=Feature_{i}^{T}^{ — }^{O1}/(Feature_{i}^{T}^{ — }^{O1}+Feature_{i}^{T}^{ — }^{O2}), - [0067]Second feature proportion: F4
_{i}^{T}^{ — }^{O1-T}^{ — }^{02}=Feature_{i}^{T}^{ — }^{O2}/(Feature_{i}^{T}^{ — }^{O1}+Feature_{i}^{T}^{ — }^{O2}), or - [0068]Absolute difference proportion: F5
_{i}^{T}^{ — }^{O1-T}^{ — }^{02}=|Feature_{i}^{T}^{ — }^{O1}−Feature_{i}^{T}^{ — }^{O2}|/(Feature_{i}^{T}^{ — }^{O1}+Feature_{i}^{T}^{ — }^{O2}). - [0069]Those having ordinary skill in the art should recognize that other forms of contrast feature calculation rules could be used and they are within the scope of the invention.
- [0070]In another embodiment of the invention, the generalization method uses the two object set feature calculation method. For a selected target object set and multiple condition object sets, it simply takes the union of all condition object sets as one condition object set to derive the spatial mapping features between multiple object sets.
- [0071]In the above feature calculation, we assume the target object set is pre-assigned. When the target object set can be flexibly assigned, the object sets can take turns to become the target object set and the features calculated accordingly. In this way, additional features are created.
- [0072]Those having ordinary skill in the art should recognize that other forms of object set combinations could be used and they are within the scope of the invention.
- [0073]II.2 Feature Learning
- [0074]After spatial mapping features are extracted, the events associated with the feature sets are determined. Practical examples of the events include “disease vs. non-disease”, “treatment vs. non-treatment”, defect types, target models, etc. The events are used as the labels for the spatial mapping features. Multiple labeled feature vectors can be used for feature learning. Feature learning includes feature selection and feature transformation.
- [0075]II.2.1 Feature Selection
- [0076]Feature selection selects a subset of features that could discriminate between pixels of different classes (labels). Feature selection reduces the dimensionality of the data to be processed by the classifier, reducing execution time and improving predictive accuracy.
- [0077]In one embodiment of the invention, the CFS (Correlation-based Feature Selection) method (Hall, M. A. “Correlation-based feature selection for discrete and numeric class machine learning.” Proceedings of the Seventeenth International Conference on Machine Learning, Stanford University, Calif. Morgan Kaufmann Publishers, 2000) is used for feature selection. The CFS algorithm is based on the following hypothesis:
- [0078]A good feature subset is one that contains features highly correlated with (predictive of) the class, yet uncorrelated with (not predictive of) each other. Other feature selection methods such as the branch and bound method, sequential forward selection method, or sequential backward selection method could also be used. (see “Devijver and Kittler, Pattern Recognition: A Statistical Approach”, PHI, NJ, 1982 PP. 212-219.”)
- [0079]II.2.2 Feature Transformation
- [0080]Feature transformation transforms the original feature set into a subset of derived features. The classifier can then be run on this derived set of features. Examples include principal components analysis, factor analysis and independent components analysis. Such techniques could result in reasonable classification performance.
- [0081]In one embodiment of the invention, the principal components analysis method is used for feature transformation. It is a dimensionality reduction technique in which d-dimensional data vectors are mapped onto vectors in a new M-dimensional space, where M<d. To do this, the mean and covariance matrix of the data are calculated, and the best M dimensional linear representation of the data is formed by the mean plus the projection of the data onto the eigenvectors of the covariance matrix corresponding to the M largest eigenvalues.
- [0082]Those having ordinary skill in the art should recognize that other feature transformation methods could be used and they are all within the scope of this invention.
- [0083]After Spatial mapping feature learning, discriminate spatial mapping features are derived. The discriminate features are used to perform spatial reasoning decision learning. In one embodiment of the invention, decision learning is achieved by a supervised learning method. It creates a classifier that assigns each target object set to one of the events.
- [0084]A number of different supervised learning methods can be used such as: decision tree, neural networks, K-nearest neighbor classifier, kernel-based classifier (such as Parzen windows), finite mixture models, Fisher's linear, Bayes quadratic, etc.
- [0085]In one embodiment of the invention, a decision tree classifier is used for decision learning. Decision trees represent a learned function for classification of discrete-valued target functions. Decision trees classify novel items by traversing the tree from the root down to a leaf node, which assigns a classification to the item (Breiman L., Friedman J. H., Olshen R. A. and Stone C. J., “Classification And Regression Trees”, Chapman &Hall/CRC, 1984, PP.18-58; Quinlan J. R., “C4.5 Programs For Machine Learning”, Morgan Kaufmann, 1993, PP.17-26; U.S. patent application Ser. No. 09/972,057 entitled, “Regulation of Hierarchic Decisions in Intelligent Systems”; U.S. patent application Ser. No. 10/081,441 entitled, “Information Integration Method for Decision Regulation in Hierarchic Decision Systems).
- [0086]In a challenging classification situation, boosting and bagging schemes can be used to enhance the results of the decision tree classifier. However, this will result in more computationally expensive decision classifiers.
- [0087]Boosting (See “Robert E. Schapire,
*The boosting approach to machine learning: An overview*. In MSRI Workshop on Nonlinear Estimation and Classification, 2002.”) is applied to improve the decision algorithm's accuracy. Boosting works by repeatedly invoking a weak learner (a learning algorithm for which the training set error rate is only slightly better than that of random guessing) on a set of training samples, over which a distribution is maintained. This distribution starts off as a uniform distribution, but is adapted so that on each iteration the weight given to incorrectly classified training examples increases. As such, the weak learner must give greater emphasis to the hard samples in the training set. - [0088]Bagging (bootstrap aggregation) (Dietterich, T. G., “
*Ensemble Methods in Machine Learning*.” In J. Kittler and F. Roli (Ed.) First International Workshop on Multiple Classifier Systems, Lecture Notes in Computer Science New York: Springer Verlag., 2000, PP. 1-15) is a very simple method of manipulating the training data set. A bootstrap sample is generated by uniformly sampling m instances from the training set with replacement. A classifier is then built from the bootstrap sample. This process is repeated several times, and the final classifier is constructed by taking a majority vote of the classifiers built using the bootstrap replicates. - [0089]Those having ordinary skill in the art should recognize that other pattern recognition and classification methods could be used and that they are within the scope of this invention.
- [0090]Spatial mapping features for the target object set can be calculated from the transformed images of the target object set and the condition object set. In one embodiment of the invention, five groups of features are calculated. Each group corresponds to one transformation of the target object set. Therefore, there are features of: no operation group, inner distance transformation group, outer distance transformation group, connected component labeling group, and zone of influence labeling group.
- [0091]This group maps the spatial relation of the no operation data of the target object set and the five transformed data of the condition object set.
- [0092](i) No Operation Target Data and No Operation Condition Data
- [0093]This collection defines the spatial mapping features associated with no operation target data and no operation condition data. If the object set in FIG. 5A is the target object set and the object set in FIG. 4A is the condition object set, the spatial relation underlying this collection is shown in FIG. 9A where the boundary of the no operation target object set is overlaid on the no operation condition object set. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0094]1. Condition ratio: P(Condition/Target), the probability of condition object set included in the target object set. This can be calculated by Area(Condition ∩ Target)/Area(Target). Where ∩ designates set intersection and Area(s) designates the area of a set s.
- [0095]2. Intersection ratio: Area(Condition ∩ Target)/Area(Condition U Target). Where U designates set union.
- [0096]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for mapping between no operation target data and no operation condition data that is within the scope of this invention.
- [0097](ii) No Operation Target Data and Inner Distance Transformed Condition Data
- [0098]This collection defines the spatial mapping features associated with no operation target data and inner distance transformed condition data. If the object set in FIG. 5A is the target object set and the object set in FIG. 4A is the condition object set, the spatial relation underlying this collection is shown in FIG. 9B where the boundary of the no operation target object set is overlaid on the inner distance transformed condition object set. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0099]1. Inner distance mean: mean of the inner distance values of the intersection pixels: (Condition ∩ Target). This indicates overlap between the two sets.
- [0100]2. Inner distance standard deviation: standard deviation of the inner distance values of the intersection pixels: (Condition ∩ Target).
- [0101]3. Normalized inner distance mean: inner distance mean divided by the mean inner distance value of the Condition object set.
- [0102]4. Normalized inner distance standard deviation: inner distance standard deviation divided by the mean inner distance value of the Condition object set.
- [0103]5. Inner distance skewness: skewness of the inner distance values of the intersection pixels: (Condition ∩ Target). Skewness is a dimensionless value.
- [0104]6. Inner distance kurtosis: kurtosis of the inner distance values of the intersection pixels: (Condition ∩ Target). Kurtosis is also a dimensionless value.
- [0105]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between no operation target data and inner distance transformed condition data that are within the scope of this invention.
- [0106](iii) No Operation Target Data and Outer Distance Transformed Condition Data
- [0107]This collection defines the spatial mapping features associated with no operation target data and outer distance transformed condition data. If the object set in FIG. 5A is the target object set and the object set in FIG. 4A is the condition object set, the spatial relation underlying this collection is shown in FIG. 9C where the boundary of the no operation target object set is overlaid on the outer distance transformed condition object set. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0108]1. Outer distance mean: mean of the outer distance values of the non-intersecting Target pixels: Target—(Condition ∩ Target).
- [0109]2. Outer distance standard deviation: standard deviation of the outer distance values of the non-intersecting Target pixels: Target—(Condition ∩ Target).
- [0110]3. Normalized outer distance mean: Outer distance mean divided by the mean outer distance of the Condition object set.
- [0111]4. Normalized outer distance standard deviation: Outer distance standard deviation divided by the mean outer distance of the Condition object set.
- [0112]5. Outer distance skewness: skewness of the non-intersecting Target pixels: Target—(Condition ∩ Target).
- [0113]6. Outer distance kurtosis: kurtosis of the non-intersecting Target pixels: Target—(Condition ∩ Target).
- [0114]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for mapping between no operation target data and outer distance transformed condition data that are within the scope of this invention.
- [0115](iv) No Operation Target Data and Connected Component Labeled Condition Data
- [0116]This collection defines the spatial mapping features associated with no operation target data and connected component labeled condition data. If the object set in FIG. 5A is the target object set and the object set in FIG. 4A is the condition object set, the spatial relation underlying this collection is shown in FIG. 9D where the boundary of the no operation target object set is overlaid on the connected component labeled condition object set. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0117]1. Intersection-component-count: Total number of Condition object set connected components in intersection pixels: (Condition ∩ Target).
- [0118]2. Intersection-component-count-ratio: Intersection-component-count divided by the total number of components in the Condition object set.
- [0119]3. Intersection-component-average-area: the average area among the intersection portion of the components having non-zero intersection pixels.
- [0120]4. Intersection-component-sd-area: the standard deviation of the areas among the intersection portion of the components having non-zero intersection pixels.
- [0121]5. Intersection-component-average-area-ratio: the average area ratio among the components having non-zero intersection pixels. Where the area ratio of a component is defined as the area of the intersection portion of the component divided by the total area of the component.
- [0122]6. Intersection-component-sd-area-ratio: the standard deviation of the area ratios among the components having non-zero intersection pixels.
- [0123]7. Intersection-component-area-entropy: The entropy value of the frequency of each unique component label within the intersection pixels p(i). That is:
$\sum _{\forall \mathrm{labeli}}^{\text{\hspace{1em}}}\ue89e\text{\hspace{1em}}\ue89e\left(P\ue8a0\left(i\right)/\mathrm{area}\right)*\mathrm{log}\ue8a0\left(p\ue8a0\left(i\right)/\mathrm{area}\right)$ - [0124]where area=Area(Condition ∩ Target) and log( ) is the 10 based logarithm function. It indicates how uniform p(i) is.
- [0125]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between no operation target data and connected component labeled condition data that are within the scope of this invention.
- [0126](v) No Operation Target Data and Zone of Influence (ZOI) Labeled Condition Data
- [0127]This collection defines the spatial mapping features associated with no operation target data and zone of influence labeled condition data. If the object set in FIG. 5A is the target object set and the object set in FIG. 4A is the condition object set, the spatial relation underlying this collection is shown in FIG. 9E where the boundary of the no operation target object set is overlaid on the zone of influence labeled condition object set. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0128]1. ZOI_intersection-component-count: Total number of Condition object set ZOI connected components within Target region.
- [0129]2. ZOI_intersection-component-count-ratio: ZOI_intersection-component-count divided by the total number of ZOI components in the Condition object set.
- [0130]3. ZOI_intersection-component-average-area: the average area among the intersection portion of the ZOI components having non-zero intersection pixels.
- [0131]4. ZOI_intersection-component-sd-area: the standard deviation of the areas among the intersection portion of the ZOI components having non-zero intersection pixels.
- [0132]5. ZOI_intersection-component-average-area-ratio: the average area ratio among the ZOI components having non-zero intersection pixels. Where the area ratio of a ZOI component is defined as the area of the intersection portion of the component divided by the total area of the component.
- [0133]6. ZOI_intersection-component-sd-area-ratio: the standard deviation of the area ratios among the ZOI components having non-zero intersection pixels.
- [0134]7. ZOI_intersection-component-area-entropy: The entropy value of the frequency of each unique ZOI component label within the intersection pixels p(i). That is:
$\sum _{\forall \mathrm{labeli}}^{\text{\hspace{1em}}}\ue89e\text{\hspace{1em}}\ue89e\left(P\ue8a0\left(i\right)/\mathrm{area}\right)*\mathrm{log}\ue8a0\left(p\ue8a0\left(i\right)/\mathrm{area}\right)$ - [0135]a. Where area=Area(Target) and log( ) is the 10 based logarithm function.
- [0136]b. It indicates how uniform p(i) is.
- [0137]8. Intersection-component-count_proportion: Intersection-component-count divided by ZOI_intersection-component-count.
- [0138]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between no operation target data and zone of influence (ZOI) labeled condition data that are within the scope of this invention.
- [0139]This group maps the spatial relation of the inner distance data of the target object set and the five transformed data of the condition object set.
- [0140](i) Inner Distance Target Data and No Operation Condition Data
- [0141]This collection defines the spatial mapping features associated with inner distance target data and no operation condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0142]1. Target-inner-distance-mean: mean of the Target object set inner distance values of the intersection pixels: (Condition ∩ Target).
- [0143]2. Target-inner-distance-standard-deviation: standard deviation of the Target object set inner distance values of the intersection pixels: (Condition ∩ Target).
- [0144]3. Normalized-target-inner-distance-mean: Target-inner-distance-mean divided by the mean inner distance value of the Target object set.
- [0145]4. Normalized-target-inner-distance-standard-deviation: Target-inner-distance-standard-deviation divided by the mean inner distance value of the Target object set.
- [0146]5. Target-inner-distance-skewness: skewness of the Target object set inner distance values of the intersection pixels: (Condition ∩ Target).
- [0147]6. Target-inner-distance-kurtosis: kurtosis of the Target object set inner distance values of the intersection pixels: (Condition ∩ Target). Kurtosis is also a dimensionless value.
- [0148]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between inner distance target data and no operation condition data that are within the scope of this invention.
- [0149](ii) Inner Distance Target Data and Inner Distance Transformed Condition Data
- [0150]This collection defines the spatial mapping features associated with inner distance target data and inner distance transformed condition data. In one embodiment of the invention, the inner distance values of the target object set and the condition object set are limited to no more than 6. If the distance value is greater than 6, they are quantized into 6 bins. A 6×6 distance co-occurrence matrix can be created called inner_to_inner_distance matrix, I_ID(i,j). The (i,j)th entry of the matrix I_ID(i,j) corresponds to the number of pixels having target inner distance value i and condition inner distance value j where i and j are between 1 and 6. A inner_to_inner_distance probability matrix, P
_{I}_{ — }_{ID}(i,j) is defined where$\begin{array}{c}{P}_{\mathrm{I\_ID}}\ue8a0\left(i,j\right)=\frac{\mathrm{I\_ID}\ue89e\left(i,j\right)}{\sum _{i=1}^{6}\ue89e\text{\hspace{1em}}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e\mathrm{I\_ID}\ue89e\left(i,j\right)}\\ \mathrm{Let}\ue89e\text{\hspace{1em}}\ue89e{\mu}_{T}=\sum _{i=1}^{6}\ue89e\text{\hspace{1em}}\ue89ei\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{I\_ID}}\ue8a0\left(i,j\right)\ue89e\text{\hspace{1em}}\ue89e\mathrm{and}\ue89e\text{\hspace{1em}}\ue89e{\mu}_{C}=\sum _{i=1}^{6}\ue89e\text{\hspace{1em}}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89ej\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{I\_ID}}\ue8a0\left(i,j\right)\\ {\sigma}_{T}^{2}=\sum _{i=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{\left(i-{\mu}_{T}\right)}^{2}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{I\_ID}}\ue8a0\left(i,j\right)\ue89e\text{\hspace{1em}}\ue89e\mathrm{and}\ue89e\text{\hspace{1em}}\ue89e{\sigma}_{C}^{2}=\sum _{i=1}^{6}\ue89e\text{\hspace{1em}}\ue89e\sum _{j=1}^{6}\ue89e{\left(j-{\mu}_{C}\right)}^{2}\ue89e{P}_{\mathrm{I\_ID}}\ue8a0\left(i,j\right)\end{array}$ - [0151]The corresponding feature collection is defined as follows:
- [0152]
- [0153]
- [0154]
- [0155]
- [0156]
- [0157]
- [0158]
- [0159]
- [0160]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between inner distance target data and inner distance transformed condition data that are within the scope of this invention.
- [0161](iii) Inner Distance Target Data and Outer Distance Transformed Condition Data
- [0162]This collection defines the spatial mapping features associated with inner distance target data and outer distance transformed condition data. In one embodiment of the invention, the inner distance values of the target object set and the outer distance values of the condition object set are limited to no more than 6. If the distance value is greater than 6, they are quantized into 6 bins. A 6×6 distance co-occurrence matrix can be created called inner_to_outer_distance matrix, I_OD(i,j). The (i,j)th entry of the matrix I_OD(i,j) corresponds to the number of pixels having target inner distance value i and condition outer distance value j where i and j are between 1 and 6. A inner_to_outer_distance probability matrix, P
_{I}_{ — }_{OD}(i,j) is defined where${P}_{\mathrm{I\_OD}}\ue8a0\left(i,j\right)=\frac{\mathrm{I\_OD}\ue89e\left(i,j\right)}{\sum _{i=1}^{6}\ue89e\text{\hspace{1em}}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e\mathrm{I\_OD}\ue89e\left(i,j\right)}$ $\begin{array}{c}\mathrm{Let}\ue89e\text{\hspace{1em}}\ue89e{\mu}_{T}=\sum _{i=1}^{6}\ue89ei\ue89e\text{\hspace{1em}}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{I\_OD}}\ue8a0\left(i,j\right)\ue89e\text{\hspace{1em}}\ue89e\mathrm{and}\ue89e\text{\hspace{1em}}\ue89e{\mu}_{c}=\sum _{i=1}^{6}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{\mathrm{jP}}_{\mathrm{I\_OD}}\ue8a0\left(i,j\right)\\ {\sigma}_{T}^{2}=\sum _{i=1}^{6}\ue89e{\left(i-{\mu}_{T}\right)}^{2}\ue89e\sum _{j=1}^{6}\ue89e{P}_{\mathrm{I\_OD}}\ue8a0\left(i,j\right)\ue89e\text{\hspace{1em}}\ue89e\mathrm{and}\ue89e\text{\hspace{1em}}\ue89e{\sigma}_{c}^{2}=\sum _{i=1}^{6}\ue89e\sum _{j=1}^{6}\ue89e{\left(j-{\mu}_{c}\right)}^{2}\ue89e{P}_{\mathrm{I\_OD}}\ue8a0\left(i,j\right)\end{array}$ - [0163]The corresponding feature collection is defined as follows:
- [0164]
- [0165]
- [0166]
- [0167]
- [0168]
- [0169]
- [0170]
- [0171]
- [0172]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between inner distance target data and outer distance transformed condition data that are within the scope of this invention.
- [0173](iv) Inner Distance Target Data and Connected Component Labeled Condition Data
- [0174]This collection defines the spatial mapping features associated with inner distance target data and connected component labeled condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0175]1. Target-inner-distance-mean-standard-deviation: standard deviation over the mean i of the Target object set inner distance values of the intersection pixels for each component i: (Condition component i ∩ Target).
- [0176]2. Target-inner-distance-standard-deviation_mean: mean of the standard deviation i of the Target object set inner distance values of the intersection pixels for each component i: (Condition component i ∩ Target).
- [0177]3. Target-inner-distance-standard-deviation_sd: standard deviation of the standard deviation i of the Target object set inner distance values of the intersection pixels for each component i: (Condition component i ∩ Target).
- [0178]4. Normalized-target-inner-distance-mean-standard-deviation: Target-inner-distance-mean-standard-deviation divided by the mean inner distance value of the Target object set.
- [0179]5. Normalized-target-inner-distance-standard-deviation_mean: Target-inner-distance-standard-deviation_mean divided by the mean inner distance value of the Target object set.
- [0180]6. Normalized-target-inner-distance-standard-deviation_sd: Target-inner-distance-standard-deviation_sd divided by the mean inner distance value of the Target object set.
- [0181]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between inner distance target data and connected component labeled condition data that are within the scope of this invention.
- [0182](v) Inner Distance Target Data and Zone of Influence (ZOI) Labeled Condition Data
- [0183]This collection defines the spatial mapping features associated with inner distance target data and zone of influence labeled condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0184]1. ZOI_target-inner-distance-mean-standard-deviation: standard deviation over the mean i of the Target object set inner distance values of the Condition object set ZOI connected component i within Target region.
- [0185]2. ZOI_target-inner-distance-standard-deviation_mean: mean of the standard deviation i of the Target object set inner distance values of the Condition object set ZOI connected components within Target region.
- [0186]3. ZOI_target-inner-distance-standard-deviation_sd: standard deviation of the standard deviation i of the Target object set inner distance values of the Condition object set ZOI connected components within Target region.
- [0187]4. ZOI_normalized-target-inner-distance-mean-standard-deviation: ZOI_target-inner-distance-mean-standard-deviation divided by the mean inner distance value of the Target object set.
- [0188]5. ZOI_normalized-target-inner-distance-standard-deviation_mean: ZOI_target-inner-distance-standard-deviation_mean divided by the mean inner distance value of the Target object set.
- [0189]6. ZOI_normalized-target-inner-distance-standard-deviation_sd: ZOI_target-inner-distance-standard-deviation_sd divided by the mean inner distance value of the Target object set.
- [0190]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between inner distance target data and zone of influence (ZOI) labeled condition data that are within the scope of this invention.
- [0191]This group maps the spatial relation of the outer distance data of the target object set and the five transformed data of the condition object set.
- [0192](i) Outer Distance Target Data and No Operation Condition Data
- [0193]This collection defines the spatial mapping features associated with outer distance target data and no operation condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0194]1. Target-outer-distance-mean: mean of the Target object set outer distance values for the pixels in (Condition ∩ Not Target).
- [0195]2. Target-outer-distance-standard-deviation: standard deviation of the Target object set outer distance values of the pixels in (Condition ∩ Not Target).
- [0196]3. Normalized-target-outer-distance-mean: Target-outer-distance-mean divided by the mean outer distance value of the Target object set.
- [0197]4. Normalized-target-outer-distance-standard-deviation: Target-outer-distance-standard-deviation divided by the mean outer distance value of the Target object set.
- [0198]5. Target-outer-distance-skewness: skewness of the Target object set outer distance values for the pixels in (Condition ∩ Not Target).
- [0199]6. Target-outer-distance-kurtosis: kurtosis of the Target object set outer distance values for the pixels in (Condition ∩ Not Target).
- [0200]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between outer distance target data and no operation condition data that are within the scope of this invention.
- [0201](ii) Outer Distance Target Data and Inner Distance Transformed Condition Data
- [0202]This collection defines the spatial mapping features associated with outer distance target data and inner distance transformed condition data. In one embodiment of the invention, the outer distance values of the target object set and the inner distance condition object set are limited to no more than 6. If the distance value is greater than 6, they are quantized into 6 bins. A 6×6 distance co-occurrence matrix can be created called outer_to_inner_distance matrix, O_ID(i,j). The (i,j)th entry of the matrix O_ID(i,j) corresponds to the number of pixels having target outer distance value i and condition inner distance value j where i and j are between 1 and 6. A outer_to_inner_distance probability matrix, P
_{O}_{ — }_{ID}(i,j) is defined where$\begin{array}{c}{P}_{\mathrm{O\_ID}}\ue8a0\left(i,j\right)=\frac{\mathrm{O\_ID}\ue89e\left(i,j\right)}{\sum _{i=1}^{6}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e\mathrm{O\_ID}\ue89e\left(i,j\right)}\\ \mathrm{Let}\ue89e\text{\hspace{1em}}\ue89e{\mu}_{T}=\sum _{i=1}^{6}\ue89ei\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{O\_ID}}\ue8a0\left(i,j\right)\ue89e\text{\hspace{1em}}\ue89e\mathrm{and}\ue89e\text{\hspace{1em}}\ue89e{\mu}_{C}=\sum _{i=1}^{6}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89ej\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{O\_ID}}\ue8a0\left(i,j\right)\\ {\sigma}_{T}^{2}=\sum _{i=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{\left(i-{\mu}_{T}\right)}^{2}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{O\_ID}}\ue8a0\left(i,j\right)\ue89e\text{\hspace{1em}}\ue89e\mathrm{and}\ue89e\text{\hspace{1em}}\\ {\sigma}_{C}^{2}=\sum _{i=1}^{6}\ue89e\sum _{j=1}^{6}\ue89e\text{\hspace{1em}}\ue89e{\left(j-{\mu}_{C}\right)}^{2}\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{O\_ID}}\ue8a0\left(i,j\right)\end{array}$ - [0203]The corresponding feature collection is defined as follows:
- [0204]
- [0205]
- [0206]
- [0207]
- [0208]
- [0209]
- [0210]
- [0211]
- [0212]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between outer distance target data and inner distance transformed condition data that are within the scope of this invention.
- [0213](iii) Outer Distance Target Data and Outer Distance Transformed Condition Data
- [0214]This collection defines the spatial mapping features associated with outer distance target data and outer distance transformed condition data. In one embodiment of the invention, the outer distance values of the target and the condition object sets are limited to no more than 6. If the distance value is greater than 6, they are quantized into 6 bins. A 6×6 distance co-occurrence matrix can be created called outer_to_outer_distance matrix, O_OD(i,j). The (i,j)th entry of the matrix O_OD(i,j) corresponds to the number of pixels having target outer distance value i and condition outer distance value j where i and j are between 1 and 6. A outer_to_outer_distance probability matrix, P
_{O}_{ — }_{OD}(i,j) is defined where${P}_{\mathrm{O\_OD}}\ue8a0\left(i,j\right)=\frac{\mathrm{O\_OD}\ue89e\left(i,j\right)}{\sum _{i=1}^{6}\ue89e\sum _{j=1}^{6}\ue89e\mathrm{O\_OD}\ue89e\left(i,j\right)}$ $\mathrm{Let}\ue89e\text{\hspace{1em}}\ue89e{\mu}_{T}=\sum _{i=1}^{6}\ue89ei\ue89e\text{\hspace{1em}}\ue89e\sum _{j=1}^{6}\ue89e{P}_{\mathrm{O\_OD}}\ue8a0\left(i,j\right)\ue89e\text{\hspace{1em}}\ue89e\mathrm{and}\ue89e\text{\hspace{1em}}\ue89e{\mu}_{C}=\sum _{i=1}^{6}\ue89e\sum _{j=1}^{6}\ue89ej\ue89e\text{\hspace{1em}}\ue89e{P}_{\mathrm{O\_OD}}\ue8a0\left(i,j\right)$ ${\sigma}_{T}^{2}=\sum _{i=1}^{6}\ue89e{\left(i-{\mu}_{T}\right)}^{2}\ue89e\sum _{j=1}^{6}\ue89e{P}_{\mathrm{O\_OD}}\ue8a0\left(i,j\right)\ue89e\text{\hspace{1em}}\ue89e\mathrm{and}$ ${\sigma}_{C}^{2}=\sum _{i=1}^{6}\ue89e\sum _{j=1}^{6}\ue89e{\left(j-{\mu}_{C}\right)}^{2}\ue89e{P}_{\mathrm{O\_OD}}\ue8a0\left(i,j\right)$ - [0215]The corresponding feature collection is defined as follows:
- [0216]
- [0217]
- [0218]
- [0219]
- [0220]
- [0221]
- [0222]
- [0223]
- [0224]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between outer distance target data and outer distance transformed condition data that are within the scope of this invention.
- [0225](iv) Outer Distance Target Data and Connected Component Labeled Condition Data
- [0226]This collection defines the spatial mapping features associated with outer distance target data and connected component labeled condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0227]1. Target-outer-distance-mean-standard-deviation: standard deviation over the mean i of the Target object set outer distance values of the intersection pixels for each component i: (Condition component i ∩ Not Target).
- [0228]2. Target-outer-distance-standard-deviation_mean: mean of the standard deviation i of the Target object set outer distance values of the intersection pixels for each component i: (Condition component i ∩ Not Target).
- [0229]3. Target-outer-distance-standard-deviation_sd: standard deviation of the standard deviation i of the Target object set outer distance values of the intersection pixels for each component i: (Condition component i ∩ Not Target).
- [0230]4. Normalized-target-outer-distance-mean-standard-deviation: Target-outer-distance-mean-standard-deviation divided by the mean outer distance value of the Target object set.
- [0231]5. Normalized-target-outer-distance-standard-deviation_mean: Target-outer-distance-standard-deviation_mean divided by the mean outer distance value of the Target object set.
- [0232]6. Normalized-target-outer-distance-standard-deviation_sd: Target-outer-distance-standard-deviation_sd divided by the mean outer distance value of the Target object set.
- [0233]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between outer distance target data and connected component labeled condition data that are within the scope of this invention.
- [0234](v) Outer Distance Target Data and Zone of Influence (ZOI) Labeled Condition Data
- [0235]This collection defines the spatial mapping features associated with outer distance target data and zone of influence labeled condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0236]1. ZOI_target-outer-distance-mean-standard-deviation: standard deviation over the mean i of the Target object set outer distance values of the Condition object set ZOI connected component i outside Target region.
- [0237]2. ZOI_target-outer-distance-standard-deviation_mean: mean of the standard deviation i of the Target object set outer distance values of the Condition object set ZOI connected components outside Target region.
- [0238]3. ZOI_target-outer-distance-standard-deviation_sd: standard deviation of the standard deviation i of the Target object set outer distance values of the Condition object set ZOI connected components outside Target region.
- [0239]4. ZOI_normalized-target-outer-distance-mean-standard-deviation: ZOI_target-outer-distance-mean-standard-deviation divided by the mean outer distance value of the Target object set.
- [0240]5. ZOI_normalized-target-outer-distance-standard-deviation_mean: ZOI_target-outer-distance-standard-deviation_mean divided by the mean outer distance value of the Target object set.
- [0241]6. ZOI_normalized-target-outer-distance-standard-deviation_sd: ZOI_target-outer-distance-standard-deviation_sd divided by the mean outer distance value of the Target object set.
- [0242]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between outer distance target data and zone of influence (ZOI) labeled condition data that are within the scope of this invention.
- [0243]This group maps the spatial relations of the connected component labeled data of the target object set and the five transformed data of the condition object set.
- [0244](i) Connected Component Labeled Target Data and No Operation Condition Data
- [0245]This collection defines the spatial mapping features associated with connected component labeled target data and no operation condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0246]1. Intersection-target-component-count: Total number of Target object set connected components in intersection pixels: (Condition ∩ Target).
- [0247]2. Intersection-target-component-count-ratio: Intersection-target-component-count divided by the total number of components in the Target object set.
- [0248]3. Intersection-target-component-average-area: the average area among the intersection portion of the target components having non-zero intersection pixels.
- [0249]4. Intersection-target-component-sd-area: the standard deviation of the areas among the intersection portion of the target components having non-zero intersection pixels.
- [0250]5. Intersection-target-component-average-area-ratio: the average area ratio among the target components having non-zero intersection pixels. Where the area ratio of a target component is defined as the area of the intersection portion of the target component divided by the total area of the component.
- [0251]6. Intersection-target-component-sd-area-ratio: the standard deviation of the area ratios among the target components having non-zero intersection pixels.
- [0252]7. Intersection-target-component-area-entropy: The entropy value of the frequency of each unique target component label within the intersection pixels p(i). That is:
$\sum _{\forall \mathrm{labeli}}\ue89e\left(P\ue8a0\left(i\right)/\mathrm{area}\right)*\mathrm{log}\left(p\ue8a0\left(i\right)/\mathrm{area}\right)$ - [0253]Where area=Area(Condition ∩ Target) and log( ) is the 10 based logarithm function. It indicates how uniform p(i) is.
- [0254]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between connected component labeled target data and no operation condition data that are within the scope of this invention.
- [0255](ii) Connected Component Labeled Target Data and Inner Distance Transformed Condition Data
- [0256]This collection defines the spatial mapping features associated with connected component labeled target data and inner distance transformed condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0257]1. Condition-inner-distance-mean-standard-deviation: standard deviation over the mean i of the Condition object set inner distance values of the intersection pixels for each Target component i (Condition ∩ Target component i).
- [0258]2. Condition-inner-distance-standard-deviation_mean: mean of the standard deviation i of the Condition object set inner distance values of the intersection pixels for each Target component i (Condition ∩ Target component i).
- [0259]3. Condition-inner-distance-standard-deviation_sd: standard deviation of the standard deviation i of the Condition object set inner distance values of the intersection pixels for each Target component i (Condition ∩ Target component i).
- [0260]4. Normalized-condition-inner-distance-mean-standard-deviation: Condition-inner-distance-mean-standard-deviation divided by the mean inner distance value of the Condition object set.
- [0261]5. Normalized-condition-inner-distance-standard-deviation_mean: Condition-inner-distance-standard-deviation_mean divided by the mean inner distance value of the Condition object set.
- [0262]6. Normalized-condition-inner-distance-standard-deviation_sd: Condition-inner-distance-standard-deviation_sd divided by the mean inner distance value of the Condition object set.
- [0263]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between connected component labeled target data and inner distance transformed condition data that are within the scope of this invention.
- [0264](iii) Connected Component Labeled Target Data and Outer Distance Transformed Condition Data
- [0265]This collection defines the spatial mapping features associated with connected component labeled target data and outer distance transformed condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0266]1. Condition-outer-distance-mean-standard-deviation: standard deviation over the mean i of the Condition object set outer distance values of the intersection pixels for each Target component i in (Not Condition ∩ Target component i).
- [0267]2. Condition-outer-distance-standard-deviation_mean: mean of the standard deviation i of the Condition object set outer distance values of the intersection pixels for each Target component i in (Not Condition ∩ Target component i).
- [0268]3. Condition-outer-distance-standard-deviation_sd: standard deviation of the standard deviation i of the Condition object set outer distance values of the intersection pixels for each Target component i in (Not Condition ∩ Target component i).
- [0269]4. Normalized-condition-outer-distance-mean-standard-deviation: Condition-outer-distance-mean-standard-deviation divided by the mean outer distance value of the Condition object set.
- [0270]5. Normalized-condition-outer-distance-standard-deviation_mean: Condition-outer-distance-standard-deviation_mean divided by the mean outer distance value of the Condition object set.
- [0271]6. Normalized-condition-outer-distance-standard-deviation_sd: Condition-outer-distance-standard-deviation_sd divided by the mean outer distance value of the Condition object set.
- [0272]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between connected component labeled target data and outer distance transformed condition data that are within the scope of this invention.
- [0273](iv) Connected Component Labeled Target Data and Connected Component Labeled Condition Data
- [0274]This collection defines the spatial mapping features associated with connected component labeled target data and connected component labeled condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0275]1. Target_intersection-component-count_mean: mean of the number of Condition object set connected components i intersect with Target component i (for Condition ∩ Target component i≠Φ).
- [0276]2. Target_intersection-component-count_sd: the standard deviation among the number of Condition object set connected components i intersect with Target component i (for Condition ∩ Target component i≠Φ).
- [0277]3. Target_intersection-component-count-ratio_mean: Target_intersection-component-count_mean divided by the total number of components in the Condition object set.
- [0278]4. Target_intersection-component-count-ratio_sd: Target_intersection-component-count_sd divided by the total number of components in the Condition object set.
- [0279]5. Target_intersection-component-average-area_mean: mean of the average area i of Condition object set connected components intersect with Target component i (for Condition ∩ Target component i≠Φ).
- [0280]6. Target_intersection-component-average-area_sd: the standard deviation among the average area i of Condition object set connected components intersect with Target component i (for Condition ∩ Target component i≠Φ).
- [0281]7. Target_intersection-component-average-area-ratio_mean: mean of the average area ratio i of Condition object set connected components intersect with Target component i (for Condition ∩ Target component i≠Φ). Where the area ratio of a component is defined as the area of the intersection portion of the component divided by the total area of the component.
- [0282]8. Target_intersection-component-average-area-ratio_sd: the standard deviation among the average area ratio i of Condition object set connected components intersect with Target component i (for Condition ∩ Target component i≠Φ). Where the area ratio of a component is defined as the area of the intersection portion of the component divided by the total area of the component.
- [0283]9. Target_intersection-component-area-entropy_mean: mean of the entropy value k of the frequency of each unique Condition component label within the intersection pixels p(i) for Target component k where Condition ∩ Target component k≠Φ. 10. Target_intersection-component-area-entropy_sd: the standard deviation among the entropy value k of the frequency of each unique Condition component label within the intersection pixels p(i) for Target component k where Condition ∩ Target component k≠Φ.
- [0284]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between connected component labeled target data and connected component labeled condition data that are within the scope of this invention.
- [0285](v) Connected Component Labeled Target Data and Zone of Influence (ZOI) Labeled Condition Data
- [0286]This collection defines the spatial mapping features associated with connected component labeled target data and zone of influence labeled condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0287]1. Target_intersection-ZOI-component-count_mean: mean of the number of Condition object set ZOI connected components i intersect with Target component i
- [0288]2. Target_intersection-ZOI-component-count_sd: the standard deviation among the number of Condition object set ZOI connected components i intersect with Target component i
- [0289]3. Target_intersection-ZOI-component-count-ratio_mean: Target_intersection-ZOI-component-count_mean divided by the total number of ZOI components in the Condition object set.
- [0290]4. Target_intersection-ZOI-component-count-ratio_sd: Target_intersection-ZOI-component-count_sd divided by the total number of ZOI components in the Condition object set.
- [0291]5. Target_intersection-ZOI-component-average-area_mean: mean of the average area i of Condition object set ZOI connected components intersect with Target component i
- [0292]6. Target_intersection-ZOI-component-average-area_sd: the standard deviation among the average area i of Condition object set ZOI connected components intersect with Target component i
- [0293]7. Target_intersection-ZOI-component-average-area-ratio_mean: mean of the average area ratio i of Condition object set ZOI connected components intersect with Target component i. Where the area ratio of a ZOI component is defined as the area of the intersection portion of the ZOI component divided by the total area of the ZOI component.
- [0294]8. Target_intersection-ZOI-component-average-area-ratio_sd: the standard deviation among the average area ratio i of Condition object set ZOI connected components intersect with Target component i. Where the area ratio of a ZOI component is defined as the area of the intersection portion of the ZOI component divided by the total area of the ZOI component.
- [0295]9. Target_intersection-ZOI-component-area-entropy_mean: mean of the entropy value k of the frequency of each unique Condition ZOI component label within the intersection pixels p(i) for Target component k.
- [0296]10. Target_intersection-ZOI-component-area-entropy_sd: the standard deviation among the entropy value k of the frequency of each unique Condition ZOI component label within the intersection pixels p(i) for Target component k.
- [0297]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between connected component labeled target data and zone of influence (ZOI) labeled condition data that are within the scope of this invention.
- [0298]This group maps the spatial relations of the ZOI labeled data of the target object set and the five transformed data of the condition object set.
- [0299]ZOI Labeled Target Data and No Operation Condition Data
- [0300]This collection defines the spatial mapping features associated with ZOI labeled target data and no operation condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0301]1. Intersection-target-ZOI-component-count: Total number of Target object set ZOI connected components in Condition pixels.
- [0302]2. Intersection-target-ZOI-component-count-ratio: Intersection-target-ZOI-component-count divided by the total number of in the Target ZOI components.
- [0303]3. Intersection-target-ZOI-component-average-area: the average area among the condition pixel portion of the target ZOI components having non-zero intersection pixels.
- [0304]4. Intersection-target-ZOI-component-sd-area: the standard deviation of the areas among the condition pixel portion of the target ZOI components having non-zero intersection pixels.
- [0305]5. Intersection-target-ZOI-component-average-area-ratio: the average area ratio among the target ZOI components having non-zero condition pixels. Where the area ratio of a target ZOI component is defined as the area of the condition pixel portion of the target ZOI component divided by the total area of the ZOI component.
- [0306]6. Intersection-target-ZOI-component-sd-area-ratio: the standard deviation of the area ratios among the target ZOI components having non-zero condition pixels.
- [0307]7. Intersection-target-ZOI-component-area-entropy: The entropy value of the frequency of each unique target ZOI component label within the condition pixels p(i). That is:
$\sum _{\forall \mathrm{label}\ue89e\text{\hspace{1em}}\ue89ei}^{\text{\hspace{1em}}}\ue89e\text{\hspace{1em}}\ue89e\left(P\ue8a0\left(i\right)/\mathrm{area}\right)*\mathrm{log}\ue8a0\left(p\ue8a0\left(i\right)/\mathrm{area}\right)$ - [0308]Where area=Area(Condition ∩ Target ZOI) and log( ) is the 10 based logarithm function. It indicates how uniform p(i) is.
- [0309]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between ZOI labeled target data and no operation condition data that are within the scope of this invention.
- [0310](ii) ZOI Labeled Target Data and Inner Distance Transformed Condition Data
- [0311]This collection defines the spatial mapping features associated with ZOI labeled target data and inner distance transformed condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0312]1. Condition-inner-distance-mean-ZOI-standard-deviation: standard deviation over the mean i of the Condition object set inner distance values of the intersection pixels for each Target ZOI component i (Condition ∩ Target ZOI component i).
- [0313]2. Condition-inner-distance-standard-deviation-ZOI-mean: mean of the standard deviation i of the Condition object set inner distance values of the intersection pixels for each Target ZOI component i (Condition ∩ Target ZOI component i).
- [0314]3. Condition-inner-distance-standard-deviation-ZOI_sd: standard deviation of the standard deviation i of the Condition object set inner distance values of the intersection pixels for each Target ZOI component i (Condition ∩ Target ZOI component i).
- [0315]4. Normalized-condition-inner-distance-mean-ZOI-standard-deviation:
- [0316]Condition-inner-distance-mean-ZOI-standard-deviation divided by the mean inner distance value of the Condition object set.
- [0317]5. Normalized-condition-inner-distance-standard-deviation-ZOI_mean: Condition-inner-distance-standard-deviation-ZOI_mean divided by the mean inner distance value of the Condition object set.
- [0318]6. Normalized-condition-inner-distance-standard-deviation-ZOI_sd: Condition-inner-distance-standard-deviation-ZOI_sd divided by the mean inner distance value of the Condition object set.
- [0319]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between ZOI labeled target data and inner distance transformed condition data that are within the scope of this invention.
- [0320](iii) ZOI Labeled Target Data and Outer Distance Transformed Condition Data
- [0321]This collection defines the spatial mapping features associated with ZOI labeled target data and outer distance transformed condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0322]1. Condition-outer-distance-mean-ZOI-standard-deviation: standard deviation over the mean i of the Condition object set outer distance values of the intersection pixels for each Target ZOI component i in (Not Condition ∩ Target ZOI component i).
- [0323]2. Condition-outer-distance-standard-deviation-ZOI_mean: mean of the standard deviation i of the Condition object set outer distance values of the intersection pixels for each Target component i in (Not Condition ∩ Target ZOI component i).
- [0324]3. Condition-outer-distance-standard-deviation-ZOI_sd: standard deviation of the standard deviation i of the Condition object set outer distance values of the intersection pixels for each Target component i in (Not Condition ∩ Target ZOI component i).
- [0325]4. Normalized-target-outer-distance-mean-ZOI-standard-deviation: Condition-outer-distance-mean-ZOI_standard-deviation divided by the mean outer distance value of the Condition object set.
- [0326]5. Normalized-condition-outer-distance-standard-deviation-ZOI_mean: Condition-outer-distance-standard-deviation ZOI_mean divided by the mean outer distance value of the Condition object set.
- [0327]6. Normalized-condition-outer-distance-standard-deviation-ZOI_sd: Condition-outer-distance-standard-deviation-ZOI_sd divided by the mean outer distance value of the Condition object set.
- [0328]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between ZOI labeled target data and outer distance transformed condition data that are within the scope of this invention.
- [0329](iv) ZOI Labeled Target Data and Connected Component Labeled Condition Data
- [0330]This collection defines the spatial mapping features associated with ZOI labeled target data and connected component labeled condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0331]1. Target-ZOI_intersection-component-count_mean: mean of the number of Condition object set connected components i intersect with Target ZOI component i (for Condition ∩ Target ZOI component i≠Φ).
- [0332]2. Target-ZOI_intersection-component-count_sd: the standard deviation among the number of Condition object set connected components i intersect with Target ZOI component i (for Condition ∩ Target ZOI component i≠Φ).
- [0333]3. Target-ZOI_intersection-component-count-ratio_mean: Target-ZOI_intersection-component-count_mean divided by the total number of components in the Condition object set.
- [0334]4. Target-ZOI_intersection-component-count-ratio_sd: Target-ZOI_intersection-component-count_sd divided by the total number of components in the Condition object set.
- [0335]5. Target-ZOI_intersection-component-average-area_mean: mean of the average area i of Condition object set connected components intersect with Target ZOI component i (for Condition ∩ Target—ZOI ZOI component i≠Φ).
- [0336]6. Target-ZOI_intersection-component-average-area_sd: the standard deviation among the average area i of Condition object set connected components intersect with Target ZOI component i (for Condition ∩ Target ZOI component i≠Φ).
- [0337]7. Target-ZOI_intersection-component-average-area-ratio_mean: mean of the average area ratio i of Condition object set connected components intersect with Target ZOI component i (for Condition ∩ Target ZOI component i≠Φ). Where the area ratio of a component is defined as the area of the intersection portion of the component divided by the total area of the component.
- [0338]8. Target-ZOI_intersection-component-average-area-ratio_sd: the standard deviation among the average area ratio i of Condition object set connected components intersect with Target ZOI component i (for Condition ∩ Target ZOI component i≠Φ). Where the area ratio of a component is defined as the area of the intersection portion of the component divided by the total area of the component.
- [0339]9. Target-ZOI_intersection-component-area-entropy_mean: mean of the entropy value k of the frequency of each unique Condition component label within the intersection pixels p(i) for Target ZOI component k where Condition ∩ Target ZOI component k≠Φ.
- [0340]10. Target-ZOI_intersection-component-area-entropy_sd: the standard deviation among the entropy value k of the frequency of each unique Condition component label within the intersection pixels p(i) for Target ZOI component k where Condition ∩ Target ZOI component k≠Φ.
- [0341]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for the mapping between ZOI labeled target data and connected component labeled condition data that are within the scope of this invention.
- [0342](v) ZOI Labeled Target Data ZOI Labeled Condition Data
- [0343]This collection defines the spatial mapping features associated with ZOI labeled target data and zone of influence (ZOI) labeled condition data. In one embodiment of the invention, the corresponding feature collection is defined as follows:
- [0344]1. Target-ZOI_intersection-ZOI-component-count_mean: mean of the number of Condition object set ZOI connected components i intersect with Target ZOI component i
- [0345]2. Target-ZOI_intersection-ZOI-component-count_sd: the standard deviation among the number of Condition object set ZOI connected components i intersect with Target ZOI component i
- [0346]3. Target-ZOI_intersection-ZOI-component-count-ratio_mean: Target-ZOI_intersection-ZOI-component-count_mean divided by the total number of ZOI components in the Condition object set.
- [0347]4. Target-ZOI_intersection-ZOI-component-count-ratio_sd: Target-ZOI_intersection-ZOI-component-count_sd divided by the total number of ZOI components in the Condition object set.
- [0348]5. Target-ZOI_intersection-ZOI-component-average-area_mean: mean of the average area i of Condition object set ZOI connected components intersect with Target ZOI component i
- [0349]6. Target-ZOI_intersection-ZOI-component-average-area_sd: the standard deviation among the average area i of Condition object set ZOI connected components intersect with Target ZOI component i
- [0350]7. Target-ZOI_intersection-ZOI-component-average-area-ratio_mean: mean of the average area ratio i of Condition object set ZOI connected components intersect with Target ZOI component i. Where the area ratio of a ZOI component is defined as the area of the intersection portion of the ZOI component divided by the total area of the ZOI component.
- [0351]8. Target-ZOI_intersection-ZOI-component-average-area-ratio_sd: the standard deviation among the average area ratio i of Condition object set ZOI connected components intersect with Target ZOI component i. Where the area ratio of a ZOI component is defined as the area of the intersection portion of the ZOI component divided by the total area of the ZOI component.
- [0352]9. Target-ZOI_intersection-ZOI-component-area-entropy_mean: mean of the entropy value k of the frequency of each unique Condition ZOI component label within the intersection pixels p(i) for Target ZOI component k.
- [0353]10. Target-ZOI_intersection-ZOI-component-area-entropy_sd: the standard deviation among the entropy value k of the frequency of each unique Condition ZOI component label within the intersection pixels p(i) for Target ZOI component k.
- [0354]Those having ordinary skill in the art should recognize that additional mapping functions could be defined for spatial mapping feature calculations that are within the scope of this invention.
- [0355]The invention has been described herein in considerable detail in order to comply with the Patent Statutes and to provide those skilled in the art with the information needed to apply the novel principles and to construct and use such specialized components as are required. However, it is to be understood that the inventions can be carried out by specifically different equipment and devices and dimensionality and relations, and that various modifications, both as to the implementation details and detailed procedures, can be accomplished without departing from the scope of the invention itself.

Patent Citations

Cited Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|

US4852018 * | Jan 7, 1987 | Jul 25, 1989 | Trustees Of Boston University | Massively parellel real-time network architectures for robots capable of self-calibrating their operating parameters through associative learning |

US5359712 * | Feb 21, 1992 | Oct 25, 1994 | Apple Computer, Inc. | Method and apparatus for transitioning between sequences of digital information |

US6912515 * | Jun 4, 2001 | Jun 28, 2005 | Xerox Corporation | Method and system for algorithm synthesis in problem solving |

US20040128081 * | Dec 18, 2002 | Jul 1, 2004 | Herschel Rabitz | Quantum dynamic discriminator for molecular agents |

US20040193707 * | Mar 28, 2003 | Sep 30, 2004 | Microsoft Corporation | Architecture and system for location awareness |

Referenced by

Citing Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|

US7526465 * | Mar 18, 2005 | Apr 28, 2009 | Sandia Corporation | Human-machine interactions |

US8190422 * | May 20, 2008 | May 29, 2012 | George Mason Intellectual Properties, Inc. | Semantic cognitive map |

US9213743 * | Dec 14, 2012 | Dec 15, 2015 | Microsoft Technology Licensing, Llc | Mining for statistical enumerated type |

US20100042401 * | May 20, 2008 | Feb 18, 2010 | Ascoli Giorgio A | Semantic Cognitive Map |

US20120076417 * | Jun 15, 2010 | Mar 29, 2012 | Canon Kabushiki Kaisha | Image recognition method and image recognition apparatus |

US20140172887 * | Dec 14, 2012 | Jun 19, 2014 | Microsoft Corporation | Mining for statistical enumerated type |

CN102460508A * | Jun 15, 2010 | May 16, 2012 | 佳能株式会社 | Image recognition method and image recognition apparatus |

Classifications

U.S. Classification | 706/47, 706/46 |

International Classification | G06F17/00, G06N5/02, G06N5/04 |

Cooperative Classification | G06N99/005, G06K9/4638, G06N5/04, G06K9/342, G06N5/025 |

European Classification | G06N99/00L, G06N5/02K2, G06N5/04, G06K9/46A3, G06K9/34C |

Legal Events

Date | Code | Event | Description |
---|---|---|---|

Apr 9, 2003 | AS | Assignment | Owner name: LEE, SHIH-JONG J., WASHINGTON Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:OH, SEHO;REEL/FRAME:013971/0064 Effective date: 20030409 |

Apr 27, 2008 | AS | Assignment | Owner name: SVISION LLC, WASHINGTON Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:LEE, SHIH-JONG J., DR.;REEL/FRAME:020859/0642 Effective date: 20080313 |

May 30, 2008 | AS | Assignment | Owner name: DRVISION TECHNOLOGIES LLC, WASHINGTON Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:SVISION LLC;REEL/FRAME:021018/0073 Effective date: 20080527 |

Dec 1, 2010 | FPAY | Fee payment | Year of fee payment: 4 |

Dec 1, 2014 | FPAY | Fee payment | Year of fee payment: 8 |

Rotate